1
00:00:00,280 --> 00:00:05,240
Speaker 1: Now here's a highlight from coast to coast AM on iHeartRadio, Jeremy.

2
00:00:05,280 --> 00:00:10,879
Speaker 2: We already have language interpreter devices, but with AI, don't

3
00:00:10,880 --> 00:00:12,160
Speaker 2: you think it'll be even better?

4
00:00:13,800 --> 00:00:15,720
Speaker 3: Absolutely? I mean one of the things you can do

5
00:00:16,560 --> 00:00:19,680
Speaker 3: with these systems is translated very easily between languages, and

6
00:00:19,720 --> 00:00:22,320
Speaker 3: there already people talking about, you know, putting it into

7
00:00:22,480 --> 00:00:25,239
Speaker 3: simple wearable devices. It might be possible in the future

8
00:00:25,680 --> 00:00:28,520
Speaker 3: that you can have a conversation with somebody in two

9
00:00:28,640 --> 00:00:31,400
Speaker 3: languages or neither of you actually speak, you know, not

10
00:00:31,640 --> 00:00:35,760
Speaker 3: as you rather understand, and you will get simultaneous translation

11
00:00:36,360 --> 00:00:39,360
Speaker 3: in those languages, you know, directly in your ear, and

12
00:00:40,479 --> 00:00:43,360
Speaker 3: you won't have to worry about potentially learning those languages,

13
00:00:43,400 --> 00:00:45,640
Speaker 3: although you know, there's obviously drawbacks to that if people

14
00:00:45,680 --> 00:00:47,040
Speaker 3: are not actually learning language.

15
00:00:47,240 --> 00:00:51,800
Speaker 2: So just incredible technology, it really is. What do you

16
00:00:51,840 --> 00:00:55,560
Speaker 2: think of driverless vehicles? That's artificial intelligence is, isn't it?

17
00:00:55,880 --> 00:00:56,120
Speaker 1: Yeah?

18
00:00:56,760 --> 00:01:01,080
Speaker 3: Yes, driver's vehicles use various kinds of AI, and obviously

19
00:01:01,080 --> 00:01:04,000
Speaker 3: we have driver's vehicles in a few cities in the US.

20
00:01:04,040 --> 00:01:07,760
Speaker 3: In San Francisco there are these Waymo taxis. In Phoenix,

21
00:01:08,080 --> 00:01:10,760
Speaker 3: you can get taxies as well. I think from Waymo

22
00:01:10,959 --> 00:01:14,040
Speaker 3: and Cruise, which is this self driving car company. It's

23
00:01:14,040 --> 00:01:18,000
Speaker 3: owned by General Motors, but it's been fairly limited areas,

24
00:01:18,000 --> 00:01:21,000
Speaker 3: and I think part of the problem is that driving

25
00:01:21,040 --> 00:01:23,160
Speaker 3: the real world is very complex and lots of things

26
00:01:23,200 --> 00:01:25,280
Speaker 3: that can happen out on the road. They're very hard

27
00:01:25,280 --> 00:01:29,399
Speaker 3: to anticipate, and unfortunately, the way we've had to create

28
00:01:29,480 --> 00:01:32,920
Speaker 3: driver's cards so far, they've needed very specific information about

29
00:01:32,959 --> 00:01:37,880
Speaker 3: where they are, very detailed mapping, and we've tried very

30
00:01:37,880 --> 00:01:40,800
Speaker 3: hard to train them for all different kinds of scenarios,

31
00:01:40,840 --> 00:01:44,560
Speaker 3: but ultimately it's very hard to have enough data from

32
00:01:44,640 --> 00:01:46,920
Speaker 3: enough place to train them to work well in all

33
00:01:47,000 --> 00:01:49,320
Speaker 3: kinds of environments, particularly in all kinds of weather, so

34
00:01:49,360 --> 00:01:51,480
Speaker 3: they don't do as well in snow, for instance. It's

35
00:01:51,480 --> 00:01:55,080
Speaker 3: a really tricky one for driver's cars, and as a result,

36
00:01:55,120 --> 00:01:57,520
Speaker 3: they haven't really been deployed that widely yet. I think

37
00:01:57,520 --> 00:01:59,960
Speaker 3: that's going to change though over the next ten year

38
00:02:00,160 --> 00:02:01,640
Speaker 3: is I think we're going to see more and more

39
00:02:02,000 --> 00:02:05,040
Speaker 3: cities where driver's cars will be available. And I do

40
00:02:05,120 --> 00:02:08,160
Speaker 3: think that some of these newer AI techniques, the sort

41
00:02:08,200 --> 00:02:10,400
Speaker 3: of techniques that I talked about in the book, will

42
00:02:10,520 --> 00:02:14,480
Speaker 3: enable us to create self driving cars that are more capable,

43
00:02:14,480 --> 00:02:17,280
Speaker 3: because we'll start they will start to actually understand and

44
00:02:17,560 --> 00:02:20,040
Speaker 3: I use the understands they're somewhat loosely as a term,

45
00:02:20,080 --> 00:02:22,600
Speaker 3: but I think they will start to understand more about

46
00:02:22,600 --> 00:02:25,399
Speaker 3: how the world actually works, and therefore they'll be able

47
00:02:25,440 --> 00:02:28,280
Speaker 3: to cope. They'll have essentially something closer to common sense.

48
00:02:29,040 --> 00:02:31,520
Speaker 3: They'll know that if a tree falls across a road,

49
00:02:31,600 --> 00:02:33,680
Speaker 3: that you have to stop the tree, even if that

50
00:02:33,760 --> 00:02:36,160
Speaker 3: had never happened in a narrative seen before.

51
00:02:37,240 --> 00:02:40,280
Speaker 2: How about people that have been killed by driverless cars.

52
00:02:40,880 --> 00:02:42,720
Speaker 3: Yeah, I mean, I think that's terrible and I think

53
00:02:42,760 --> 00:02:45,280
Speaker 3: that's a problem with the technology we have. So far,

54
00:02:46,400 --> 00:02:50,280
Speaker 3: there haven't been too many incidants of people being killed

55
00:02:50,320 --> 00:02:53,560
Speaker 3: by one of these ROWA taxes are full self driving cars.

56
00:02:53,560 --> 00:02:57,760
Speaker 3: There was the one incident that in Phoenix, there was

57
00:02:57,760 --> 00:02:59,720
Speaker 3: another intent where somebody had already been hit by a

58
00:02:59,720 --> 00:03:01,919
Speaker 3: car than was dragged by driver's car.

59
00:03:03,040 --> 00:03:04,919
Speaker 2: Phoenix was hit by one, right.

60
00:03:05,120 --> 00:03:06,960
Speaker 3: Yeah, the woman in Phoenix was hit by one, and

61
00:03:07,000 --> 00:03:09,240
Speaker 3: that that is a case where it's that again it

62
00:03:09,280 --> 00:03:12,560
Speaker 3: was a censor. It did not recognize her as the person.

63
00:03:13,160 --> 00:03:16,320
Speaker 3: I think again that's why we need careful testing of

64
00:03:16,320 --> 00:03:19,000
Speaker 3: these systems. We have to remember that, you know, fifty

65
00:03:19,040 --> 00:03:21,680
Speaker 3: thousand Americans are killed every year by other humans on

66
00:03:21,720 --> 00:03:24,639
Speaker 3: the road, and I think we can do better than that.

67
00:03:25,560 --> 00:03:27,280
Speaker 3: You know, that would be a good thing. And I

68
00:03:27,320 --> 00:03:28,760
Speaker 3: think it's one of the things we should get used

69
00:03:28,800 --> 00:03:31,080
Speaker 3: to when we think about these systems, is you know,

70
00:03:31,080 --> 00:03:34,560
Speaker 3: what are we comparing against and what standard do we require.

71
00:03:35,560 --> 00:03:37,160
Speaker 3: We have lots of humans on the roads that are

72
00:03:37,160 --> 00:03:40,680
Speaker 3: not particularly safe drivers every day, and if we can

73
00:03:40,720 --> 00:03:43,440
Speaker 3: do better than that, should that be the standard or

74
00:03:43,480 --> 00:03:45,800
Speaker 3: should the standard be one hundred percent you know, accuracy

75
00:03:45,800 --> 00:03:46,320
Speaker 3: and no death.

76
00:03:46,440 --> 00:03:46,720
Speaker 4: Ever.

77
00:03:47,040 --> 00:03:49,280
Speaker 3: I think there is this interesting where we were willing

78
00:03:49,320 --> 00:03:54,200
Speaker 3: to countenance certain kinds of human error, that including fatal

79
00:03:54,280 --> 00:03:57,840
Speaker 3: human errors, and think that that's okay or just the

80
00:03:57,840 --> 00:04:01,280
Speaker 3: price of life. And yet you're not willing to put

81
00:04:01,360 --> 00:04:05,240
Speaker 3: systems into place, automated systems that would actually be less

82
00:04:05,320 --> 00:04:07,960
Speaker 3: dangerous than having humans do it. Because I think that

83
00:04:08,040 --> 00:04:09,720
Speaker 3: part of it has to do we don't really understand

84
00:04:09,880 --> 00:04:13,120
Speaker 3: the way these errors occur when other humans make mistakes.

85
00:04:13,320 --> 00:04:14,920
Speaker 3: We kind of have a good sense of what the

86
00:04:14,960 --> 00:04:18,080
Speaker 3: things are that cause humans to make mistakes are. What

87
00:04:18,120 --> 00:04:22,000
Speaker 3: are the human errors that can happen to cause to cause,

88
00:04:22,520 --> 00:04:25,040
Speaker 3: you know, fatal accidents. We don't have a good sense

89
00:04:25,080 --> 00:04:27,080
Speaker 3: of exactly what all the errors are that can crop

90
00:04:27,200 --> 00:04:30,200
Speaker 3: up in these automated AI systems, and I think that

91
00:04:30,200 --> 00:04:32,159
Speaker 3: that frightens us a bit more, and we're a little

92
00:04:32,160 --> 00:04:35,600
Speaker 3: bit more hesitant to put these systems into places of

93
00:04:35,680 --> 00:04:38,559
Speaker 3: high consequence. And I think, you know, that's that's that's good.

94
00:04:38,600 --> 00:04:40,560
Speaker 3: I think we should be careful, but I think we

95
00:04:40,600 --> 00:04:43,800
Speaker 3: should also have a standard. What are we comparing this against?

96
00:04:44,320 --> 00:04:45,760
Speaker 3: What do we want these systems to do.

97
00:04:46,640 --> 00:04:49,960
Speaker 2: I'm opposed to driverless trucks, jure, maybe because I don't

98
00:04:49,960 --> 00:04:53,160
Speaker 2: want to see truck drivers displaced from jobs.

99
00:04:54,080 --> 00:04:57,240
Speaker 3: Yeah, well, look, I mean I think that it is

100
00:04:57,279 --> 00:05:01,039
Speaker 3: potentially an issue, but there are of other jobs people

101
00:05:01,040 --> 00:05:02,880
Speaker 3: can do. I think we don't have to worry so

102
00:05:02,960 --> 00:05:05,080
Speaker 3: much about math employment. They're all going to with their jobs.

103
00:05:05,120 --> 00:05:07,560
Speaker 3: But if you think in certain industries we may see

104
00:05:07,960 --> 00:05:09,920
Speaker 3: people lose their jobs. I think in case a trucking

105
00:05:10,000 --> 00:05:12,560
Speaker 3: is really a shortage of trucking driver. So the question is,

106
00:05:12,839 --> 00:05:15,240
Speaker 3: you know, does this mean that all truckers you know,

107
00:05:15,279 --> 00:05:16,839
Speaker 3: lose their jobs or does it just mean that some

108
00:05:16,960 --> 00:05:19,880
Speaker 3: trucks are automated and there's still a role for human drivers.

109
00:05:20,000 --> 00:05:22,039
Speaker 3: I think it depends. That depends what the cost is

110
00:05:22,080 --> 00:05:26,120
Speaker 3: of these systems and how they're deployed exactly. It's one

111
00:05:26,120 --> 00:05:28,000
Speaker 3: of the things we're going to have to wrestle with

112
00:05:28,080 --> 00:05:28,640
Speaker 3: going forward.

113
00:05:28,960 --> 00:05:31,040
Speaker 2: Let's go to the phones. Let's go to Thomas and

114
00:05:31,160 --> 00:05:35,440
Speaker 2: La Jolla, California to start. Hey, Tom, welcome, Hi George,

115
00:05:35,560 --> 00:05:38,840
Speaker 2: Thank you very much. I'm getting a lot of static

116
00:05:38,920 --> 00:05:41,960
Speaker 2: on my line, so I will try to make the

117
00:05:42,080 --> 00:05:49,000
Speaker 2: short I want to extend the conversation on self driving trucks.

118
00:05:49,920 --> 00:05:56,320
Speaker 2: In my opinion, it will not displice jobs. Okay, if

119
00:05:56,360 --> 00:05:59,839
Speaker 2: we have a truck that is automated, it's going to

120
00:05:59,839 --> 00:06:04,760
Speaker 2: be like the Apollo Moon capsule, you know, with AI

121
00:06:05,600 --> 00:06:08,760
Speaker 2: running the truck. But there will always be a teamster

122
00:06:09,640 --> 00:06:13,960
Speaker 2: there in the truck always, and they will be like astronauts.

123
00:06:14,520 --> 00:06:16,800
Speaker 2: You know, They're not going to be using a steering

124
00:06:16,880 --> 00:06:20,480
Speaker 2: wheel or a stick shift, or a brake pedal or

125
00:06:20,640 --> 00:06:24,599
Speaker 2: gas pedal, or a rear view mirror or any of that.

126
00:06:25,160 --> 00:06:28,560
Speaker 2: You know, it's all going to be handled by the AI.

127
00:06:29,400 --> 00:06:35,839
Speaker 2: But you know, occasionally the AI will say, sir or madam,

128
00:06:36,839 --> 00:06:40,719
Speaker 2: we need to make a course correction, and so our

129
00:06:40,760 --> 00:06:47,120
Speaker 2: teamster puts down his Fortune magazine and participates in the

130
00:06:47,839 --> 00:06:51,560
Speaker 2: course correction. So I was wondering, it just seems to

131
00:06:51,600 --> 00:06:55,560
Speaker 2: me we should have this technology available now. We had

132
00:06:55,560 --> 00:06:59,680
Speaker 2: it available like in the Apollo Moon Capsule fifty five

133
00:06:59,760 --> 00:07:04,880
Speaker 2: years ago, and the Apollo Moon Capsule only used a

134
00:07:04,960 --> 00:07:10,480
Speaker 2: computer that was two megabytes two megabytes, and of course

135
00:07:10,480 --> 00:07:15,240
Speaker 2: our smartphones have hundreds of gigabytes. So I'm wondering if

136
00:07:15,240 --> 00:07:18,400
Speaker 2: you would if we could extend the conversation. I don't

137
00:07:18,400 --> 00:07:22,960
Speaker 2: think we'll lose jobs. I think that our truck drivers

138
00:07:23,000 --> 00:07:24,920
Speaker 2: will be trained like astronauts.

139
00:07:25,200 --> 00:07:28,720
Speaker 4: What do you think, sir, I.

140
00:07:28,640 --> 00:07:31,280
Speaker 3: Mean, I think yeah, I think that's quite That's a

141
00:07:31,560 --> 00:07:34,800
Speaker 3: possible scenario. People have also talked about whether you would

142
00:07:34,800 --> 00:07:37,360
Speaker 3: have still the truckers, but they would operate in some

143
00:07:37,400 --> 00:07:39,960
Speaker 3: sort of remote center and you would have AI Infensa

144
00:07:40,080 --> 00:07:42,000
Speaker 3: driving the trucks most of the time, and then if

145
00:07:42,000 --> 00:07:44,520
Speaker 3: there was a problem, it would alert the driver, but

146
00:07:44,560 --> 00:07:47,800
Speaker 3: the driver wouldn't necessarily be in the truck themselves. They

147
00:07:47,880 --> 00:07:50,040
Speaker 3: might be in a remote center where they could take

148
00:07:50,040 --> 00:07:53,440
Speaker 3: over a remote operation of the truck using off the camera.

149
00:07:53,480 --> 00:07:55,280
Speaker 3: They'd be able to see what's going on and they

150
00:07:55,680 --> 00:07:58,360
Speaker 3: would then drive the trucks that are remotely if they're

151
00:07:58,360 --> 00:08:00,160
Speaker 3: an error, at least the side of the road, and

152
00:08:00,200 --> 00:08:02,520
Speaker 3: then you could have a recovery vehicle coming and and

153
00:08:02,640 --> 00:08:05,880
Speaker 3: move the truck or yeah, I mean as possible as

154
00:08:05,920 --> 00:08:08,200
Speaker 3: you say, we will want the teacher or the truck

155
00:08:08,280 --> 00:08:11,200
Speaker 3: driver to actually be in the cab, but they'll just

156
00:08:11,280 --> 00:08:14,800
Speaker 3: have to sit there and wait until there's a problem

157
00:08:14,880 --> 00:08:17,920
Speaker 3: or something that needs human intervention or human decision making.

158
00:08:18,600 --> 00:08:22,400
Speaker 3: That's that is one possible scenario. I think that that

159
00:08:22,600 --> 00:08:26,160
Speaker 3: could happen. It's some of the company that are pushing

160
00:08:26,200 --> 00:08:28,720
Speaker 3: self driving trucks though, really do see it as a

161
00:08:29,000 --> 00:08:31,720
Speaker 3: system where there is no driver there at least for

162
00:08:31,840 --> 00:08:33,520
Speaker 3: large stretches of the journey.

163
00:08:33,360 --> 00:08:38,240
Speaker 2: They're doing they're doing it money, aren't they.

164
00:08:39,080 --> 00:08:40,160
Speaker 4: I'm sorry said that again, George.

165
00:08:40,240 --> 00:08:43,000
Speaker 2: They're doing it to save money. There won't be a truck.

166
00:08:43,040 --> 00:08:45,040
Speaker 3: Well, yeah, they want that is right. They want to

167
00:08:45,040 --> 00:08:48,839
Speaker 3: save money. They but they also want to they want

168
00:08:48,840 --> 00:08:51,120
Speaker 3: to save money and they also want to In a

169
00:08:51,160 --> 00:08:53,400
Speaker 3: case of trucking, there aren't enough truck drivers to actually

170
00:08:53,440 --> 00:08:55,880
Speaker 3: haul all the things that be hauling, and there's a

171
00:08:55,880 --> 00:08:58,240
Speaker 3: real labor short Not enough people are going into trucking.

172
00:08:58,400 --> 00:09:01,040
Speaker 3: So this is a case where you know there aren't

173
00:09:01,120 --> 00:09:04,520
Speaker 3: enough people doing these jobs. And I think AI could

174
00:09:04,520 --> 00:09:08,120
Speaker 3: actually help because you could have some self driving trucks.

175
00:09:08,200 --> 00:09:09,640
Speaker 3: You could have them in a convoy where you have

176
00:09:09,679 --> 00:09:12,320
Speaker 3: one human driver to front, all the others essentially follow

177
00:09:12,360 --> 00:09:17,199
Speaker 3: and are driven automatically. I think those scenarios, you know,

178
00:09:17,440 --> 00:09:20,840
Speaker 3: could help us with the shortage of truck drivers that

179
00:09:20,880 --> 00:09:21,080
Speaker 3: we have.

180
00:09:21,920 --> 00:09:23,880
Speaker 2: I may have been the last reporter to talk to

181
00:09:24,360 --> 00:09:29,120
Speaker 2: Jimmy Hoffa back in nineteen seventy five. Jeremy, he would

182
00:09:29,200 --> 00:09:36,200
Speaker 2: be going ballistic right now. Yeah, driver trucks.

183
00:09:36,679 --> 00:09:38,520
Speaker 3: I think that's true, and a lot of other industries,

184
00:09:38,559 --> 00:09:41,040
Speaker 3: you know, people have gone ballistic. Unions have been very

185
00:09:41,120 --> 00:09:45,320
Speaker 3: upset about displacement of workers, but in the end, I

186
00:09:45,360 --> 00:09:47,840
Speaker 3: think the people have found those jobs.

187
00:09:47,840 --> 00:09:48,080
Speaker 2: There.

188
00:09:48,320 --> 00:09:50,360
Speaker 3: Again, I don't think we're gonna have massive employment. There

189
00:09:50,400 --> 00:09:54,160
Speaker 3: will be jobs for people to do, and I think ultimately,

190
00:09:54,679 --> 00:09:57,880
Speaker 3: you know, this technology is coming along, and I think

191
00:09:57,960 --> 00:09:59,319
Speaker 3: you have to figure out how are you going to

192
00:09:59,360 --> 00:10:02,719
Speaker 3: work along outside it? And are the ways to shape it?

193
00:10:02,800 --> 00:10:05,400
Speaker 3: And I think you know, one of possibility is, as

194
00:10:05,400 --> 00:10:07,960
Speaker 3: the caller suggested, that you have truck drivers in the cab,

195
00:10:09,080 --> 00:10:11,839
Speaker 3: but they are doing less of the actual driving. They're

196
00:10:11,880 --> 00:10:14,480
Speaker 3: kind of supervising the automated system. In the book, I

197
00:10:14,480 --> 00:10:16,439
Speaker 3: talk a lot about some problems. I actually use the

198
00:10:16,480 --> 00:10:18,440
Speaker 3: examples of Massa. Massa is on a ton of research

199
00:10:18,440 --> 00:10:21,120
Speaker 3: on this, because astern outs are very much often in

200
00:10:21,120 --> 00:10:24,360
Speaker 3: this role of supervising automated systems, and I think that's

201
00:10:24,440 --> 00:10:26,160
Speaker 3: increasingly going to be all of us. We're going to

202
00:10:26,200 --> 00:10:28,839
Speaker 3: be in our jobs doing a lot more supervising of

203
00:10:29,320 --> 00:10:32,520
Speaker 3: AI than we are maybe doing the task ourselves. The

204
00:10:32,559 --> 00:10:35,080
Speaker 3: problems is that humans are not very good at being

205
00:10:35,200 --> 00:10:38,520
Speaker 3: that vigilant over long periods of time. So a pretty

206
00:10:38,520 --> 00:10:40,720
Speaker 3: bu if the system is pretty good and it doesn't

207
00:10:40,720 --> 00:10:44,040
Speaker 3: make mistakes very often, it's very hard to maintain that vigilance,

208
00:10:44,480 --> 00:10:48,000
Speaker 3: and often we fail to catch the errors when they happen.

209
00:10:48,080 --> 00:10:51,320
Speaker 3: And there's this human cognitive bias that I talked about

210
00:10:51,360 --> 00:10:53,679
Speaker 3: in the book called automation bias, where we tend to

211
00:10:53,760 --> 00:10:56,840
Speaker 3: assume that the automated system is right, even in the

212
00:10:56,880 --> 00:10:59,520
Speaker 3: face of contradictory data that should alert us to the

213
00:10:59,520 --> 00:11:02,360
Speaker 3: fact that the system is making a mistake. Very often

214
00:11:02,400 --> 00:11:05,120
Speaker 3: people think, oh, no, the system, you know, it's a computer.

215
00:11:05,160 --> 00:11:08,319
Speaker 3: It must know best, even when what it's suggesting kind

216
00:11:08,320 --> 00:11:10,760
Speaker 3: of defies common sense. And I think that's a danger

217
00:11:10,760 --> 00:11:12,319
Speaker 3: we're going to have to guard against. And the other

218
00:11:12,320 --> 00:11:14,600
Speaker 3: thing that happens with people when they're placing that role

219
00:11:14,640 --> 00:11:18,200
Speaker 3: of kind of just being this the overseer of what

220
00:11:18,600 --> 00:11:21,680
Speaker 3: an automated system does is when they when a system

221
00:11:21,720 --> 00:11:24,440
Speaker 3: does go wrong, if they do recognize the air that

222
00:11:24,480 --> 00:11:27,280
Speaker 3: there is in there, they're often quite surprised by the error.

223
00:11:27,679 --> 00:11:30,280
Speaker 3: They have a lot of trouble figuring out what's going

224
00:11:30,320 --> 00:11:34,480
Speaker 3: wrong exactly and taking corrective action. That's increasingly been a

225
00:11:34,480 --> 00:11:36,480
Speaker 3: problem in aviation, where we have, again a lot of

226
00:11:36,480 --> 00:11:39,240
Speaker 3: automated systems, a lot of autopilots. Most of the time

227
00:11:39,320 --> 00:11:42,280
Speaker 3: they work perfectly well. When they do go wrong, pilots

228
00:11:42,320 --> 00:11:46,000
Speaker 3: often struggle because they're surprised that there's an error. They

229
00:11:46,040 --> 00:11:49,240
Speaker 3: often kind of panic and will make mistakes in the

230
00:11:49,280 --> 00:11:52,240
Speaker 3: process of trying to correct and will work back to

231
00:11:52,280 --> 00:11:55,360
Speaker 3: a manual process. And the key, as NASA's found and

232
00:11:55,559 --> 00:11:57,800
Speaker 3: as we found in aviation, is very often just to

233
00:11:57,920 --> 00:12:02,679
Speaker 3: drill people in simulation with lots of potential error scenarios

234
00:12:03,160 --> 00:12:05,200
Speaker 3: so they know how to respond and they're kind of

235
00:12:05,200 --> 00:12:07,760
Speaker 3: practice and how to respond. And I think companies are

236
00:12:07,760 --> 00:12:09,520
Speaker 3: going to have to start doing this as well as

237
00:12:09,520 --> 00:12:12,439
Speaker 3: we move with more of these AI copilots into how

238
00:12:12,440 --> 00:12:14,840
Speaker 3: we work, and maybe that we'll all have to train.

239
00:12:14,920 --> 00:12:17,960
Speaker 3: If you're the salesperson and you often rely on the

240
00:12:18,080 --> 00:12:21,520
Speaker 3: AI copidits to write your sales pitch for you, companies

241
00:12:21,520 --> 00:12:23,760
Speaker 3: may have to say once a month do a drill

242
00:12:23,800 --> 00:12:25,400
Speaker 3: where you have to write the sales pitch on your

243
00:12:25,400 --> 00:12:28,600
Speaker 3: own to just so you keep those human skills and

244
00:12:28,640 --> 00:12:31,520
Speaker 3: you know how to recover if the AI system isn't available.

245
00:12:32,120 --> 00:12:34,520
Speaker 2: Go to my condenver Michael, go ahead.

246
00:12:35,720 --> 00:12:38,160
Speaker 4: George, thank you for taking my callage. So great to

247
00:12:38,240 --> 00:12:40,760
Speaker 4: be with you again. And Jeremy, thank you for this

248
00:12:40,960 --> 00:12:44,719
Speaker 4: incredible presentation tonight. And what I was going to ask

249
00:12:44,760 --> 00:12:49,720
Speaker 4: you about. Interesting kind of headline here, Dee Appliances is

250
00:12:49,840 --> 00:12:53,480
Speaker 4: using Google Cloud AI to make recipes from what's already

251
00:12:53,520 --> 00:12:56,680
Speaker 4: inside of fridge. And you know, this is something that's

252
00:12:56,800 --> 00:12:59,520
Speaker 4: a relatively new thing, something I don't think AI could

253
00:12:59,520 --> 00:13:02,760
Speaker 4: do at conception. And that leads me to my question,

254
00:13:02,880 --> 00:13:05,640
Speaker 4: which is what are some misconceptions when it comes to

255
00:13:06,520 --> 00:13:09,920
Speaker 4: say what businesses, for example, think AI can do as

256
00:13:09,960 --> 00:13:14,800
Speaker 4: far as like AIS capabilities and what AI AI actually

257
00:13:14,880 --> 00:13:15,319
Speaker 4: can do.

258
00:13:17,440 --> 00:13:20,480
Speaker 3: Yeah, so this is a great a great question. Uh,

259
00:13:20,520 --> 00:13:21,960
Speaker 3: And it's a great example. I mean, I think it's

260
00:13:22,000 --> 00:13:24,080
Speaker 3: one of the things that the AI systems that have

261
00:13:24,160 --> 00:13:26,240
Speaker 3: kind of come online in the last two years can do.

262
00:13:26,559 --> 00:13:28,360
Speaker 3: Is like it could take you could just take a

263
00:13:28,400 --> 00:13:32,440
Speaker 3: picture of your fridge actually now and you you will

264
00:13:32,480 --> 00:13:35,480
Speaker 3: recognize what those items are in your fridge, and then

265
00:13:35,559 --> 00:13:37,840
Speaker 3: you know, it will suggest recipes based on those items,

266
00:13:37,840 --> 00:13:39,960
Speaker 3: which is pretty amazing. And then I think the next

267
00:13:39,960 --> 00:13:42,439
Speaker 3: step that we're about to see, uh, which I talked

268
00:13:42,440 --> 00:13:44,520
Speaker 3: about the beginning of the program, are these AI agents

269
00:13:44,720 --> 00:13:46,880
Speaker 3: And in that case, it would would actually recognize what

270
00:13:46,960 --> 00:13:49,920
Speaker 3: was in your fridge. You could suggest recipes for tonight,

271
00:13:50,000 --> 00:13:52,520
Speaker 3: but it would also know, oh, you're running low on

272
00:13:52,640 --> 00:13:55,920
Speaker 3: milk or you're out of uh, you know this particular ingredient,

273
00:13:55,960 --> 00:13:58,200
Speaker 3: and it would go online and order those things for you.

274
00:13:58,240 --> 00:14:00,000
Speaker 3: It might be your kind of weekly shop for you

275
00:14:00,120 --> 00:14:03,120
Speaker 3: online without you having to tell it very much. It

276
00:14:03,200 --> 00:14:05,320
Speaker 3: might learn your preferences and just be able to do

277
00:14:05,360 --> 00:14:08,120
Speaker 3: this automatically, which and I think that's coming in terms

278
00:14:08,120 --> 00:14:10,280
Speaker 3: of what AI can do right now. And there is

279
00:14:10,320 --> 00:14:13,520
Speaker 3: sometimes a disconnect particularly, there's a disconnect in large companies

280
00:14:13,520 --> 00:14:16,360
Speaker 3: I think between what the CEOs or the board think

281
00:14:16,400 --> 00:14:18,840
Speaker 3: AI can do, and then if you talk to the

282
00:14:18,840 --> 00:14:21,280
Speaker 3: actual engineers who have to create this technology, they're often

283
00:14:21,320 --> 00:14:24,160
Speaker 3: sort of frustrated by what the capabilities are. This often

284
00:14:24,160 --> 00:14:26,840
Speaker 3: comes down to reliability. So a lot of AI systems

285
00:14:26,880 --> 00:14:29,440
Speaker 3: right now can do lots of things, some of the

286
00:14:29,560 --> 00:14:32,320
Speaker 3: time very well, and then some of the times, on

287
00:14:32,720 --> 00:14:35,560
Speaker 3: what seems like a very similar task, or it's the

288
00:14:35,560 --> 00:14:39,360
Speaker 3: same task but you've given the instruction slightly differently, the

289
00:14:39,400 --> 00:14:42,920
Speaker 3: system will fail completely. And I think that's frustrating for people.

290
00:14:43,600 --> 00:14:45,560
Speaker 3: This is something of all the tech companies that are

291
00:14:45,600 --> 00:14:47,720
Speaker 3: working on AI systems are working very hard on is

292
00:14:47,760 --> 00:14:50,680
Speaker 3: to increase the reliability of these systems. There's some debate

293
00:14:50,800 --> 00:14:53,120
Speaker 3: in the field about how easy it is to do this,

294
00:14:53,240 --> 00:14:55,760
Speaker 3: how whether this will be possible. Some people think the

295
00:14:56,240 --> 00:15:00,400
Speaker 3: underlying large language models that we use right now create

296
00:15:00,440 --> 00:15:03,320
Speaker 3: AI software that they have a kind of fundamental problem

297
00:15:03,320 --> 00:15:05,840
Speaker 3: where they will never get to the kind of accuracy

298
00:15:06,280 --> 00:15:08,960
Speaker 3: that we need them to to really be reliable in

299
00:15:09,000 --> 00:15:10,800
Speaker 3: these systems, and that we're going to need something else,

300
00:15:11,360 --> 00:15:15,000
Speaker 3: some other kind of change in the architecture to improve

301
00:15:15,040 --> 00:15:17,800
Speaker 3: these systems. One would be, as I talked about, trying

302
00:15:17,800 --> 00:15:20,440
Speaker 3: to do more training and simulation of AI systems, So

303
00:15:20,480 --> 00:15:22,280
Speaker 3: you put them in that kind of state environment, but

304
00:15:22,320 --> 00:15:25,000
Speaker 3: let them just experiment and learn through trial and error.

305
00:15:25,240 --> 00:15:28,320
Speaker 3: That does tend to result in more reliable AI systems,

306
00:15:28,320 --> 00:15:30,760
Speaker 3: and I think it's one approach that we may see

307
00:15:30,840 --> 00:15:35,480
Speaker 3: companies pursue going forward. Some specific things I think they're

308
00:15:35,600 --> 00:15:38,560
Speaker 3: very good at right now. I think in any sort

309
00:15:38,600 --> 00:15:43,920
Speaker 3: of scenario which involves composing something that is going to

310
00:15:43,920 --> 00:15:46,160
Speaker 3: be then checked by a human, So if it's a

311
00:15:46,200 --> 00:15:50,119
Speaker 3: particularly any kind of text they're writing, writing letters, writing documents,

312
00:15:50,280 --> 00:15:52,240
Speaker 3: as only kind of a human in a loop in

313
00:15:52,280 --> 00:15:53,960
Speaker 3: that process, and it's one of the reasons we may

314
00:15:54,000 --> 00:15:56,400
Speaker 3: not see much job loss from AIS. I think you

315
00:15:56,440 --> 00:15:57,800
Speaker 3: still are going to need this human in the loop

316
00:15:58,200 --> 00:16:00,640
Speaker 3: anytime in the near term to check what the AI

317
00:16:00,720 --> 00:16:01,160
Speaker 3: is doing.

318
00:16:01,520 --> 00:16:02,480
Speaker 4: And in those.

319
00:16:02,280 --> 00:16:04,360
Speaker 3: Scenarios it's probably states because if a human check on

320
00:16:04,400 --> 00:16:06,480
Speaker 3: what's being put out and ultimately it's just a kind

321
00:16:06,520 --> 00:16:11,240
Speaker 3: of a documentation, it's not directly taking action. So those

322
00:16:11,240 --> 00:16:14,720
Speaker 3: are safer scenarios. I think situations where you would see AI,

323
00:16:15,000 --> 00:16:17,560
Speaker 3: you know, actually taking action, these kind of AI agents

324
00:16:17,560 --> 00:16:19,840
Speaker 3: that I think are coming along those I'm a little

325
00:16:19,840 --> 00:16:22,480
Speaker 3: bit more worried about given the current reliability of systems,

326
00:16:22,520 --> 00:16:25,760
Speaker 3: because they are error prone and it will be harder

327
00:16:25,800 --> 00:16:27,800
Speaker 3: to have a human check on what they're doing, I think.

328
00:16:28,560 --> 00:16:30,920
Speaker 3: And that's the case where I think businesses, you know,

329
00:16:31,240 --> 00:16:33,120
Speaker 3: have to be careful not to run ahead of themselves

330
00:16:33,120 --> 00:16:35,200
Speaker 3: and think, oh, we can have an AI system that

331
00:16:35,240 --> 00:16:39,360
Speaker 3: will automatically do our purchasing for us, will automatically do

332
00:16:39,440 --> 00:16:42,160
Speaker 3: a particular process within in the company. Maybe it's moving

333
00:16:42,640 --> 00:16:45,440
Speaker 3: data even from like one one type of software to

334
00:16:45,480 --> 00:16:48,280
Speaker 3: another often that you know people want to use AI

335
00:16:48,360 --> 00:16:51,760
Speaker 3: could do that to kind of transfer data between different programs.

336
00:16:52,560 --> 00:16:54,320
Speaker 3: That again, you're going to have to be very careful

337
00:16:54,360 --> 00:16:56,400
Speaker 3: that it's not making mistakes in the process.

338
00:16:56,960 --> 00:17:00,480
Speaker 2: Surely, if you asked AI what is God? Would it

339
00:17:00,520 --> 00:17:01,320
Speaker 2: give you an answer?

340
00:17:02,480 --> 00:17:04,239
Speaker 3: Yeah, absolutely, it would give you an answer. It might,

341
00:17:04,480 --> 00:17:06,119
Speaker 3: you know if that might, it doesn't. But you have

342
00:17:06,119 --> 00:17:08,679
Speaker 3: to remember that right now, these systems do they just

343
00:17:08,720 --> 00:17:10,600
Speaker 3: make up an answer, right, So it would give you

344
00:17:10,600 --> 00:17:12,199
Speaker 3: an answer. It would give you an answer based on

345
00:17:13,080 --> 00:17:15,160
Speaker 3: you know what the data has been trained on, which

346
00:17:15,240 --> 00:17:17,719
Speaker 3: is mostly human written data about you know what if

347
00:17:17,800 --> 00:17:20,240
Speaker 3: humans in the past, then what is God? It would

348
00:17:20,359 --> 00:17:22,280
Speaker 3: and it would probably give you whatever the kind of

349
00:17:22,320 --> 00:17:25,760
Speaker 3: majority viewpoint expressed, and that data is within some sort

350
00:17:25,800 --> 00:17:28,520
Speaker 3: of range, and it might give you several different opinions.

351
00:17:28,520 --> 00:17:29,960
Speaker 3: It might say, oh, some people think God is this,

352
00:17:30,080 --> 00:17:31,640
Speaker 3: and some people think God is that, and some people

353
00:17:31,680 --> 00:17:34,600
Speaker 3: think God doesn't exist. It would definitely give you an answer,

354
00:17:34,680 --> 00:17:37,919
Speaker 3: but that doesn't mean it's the correct answer. You know, AI,

355
00:17:38,000 --> 00:17:40,959
Speaker 3: I think pretty AI systems right now they're not omnissioned.

356
00:17:41,040 --> 00:17:44,120
Speaker 3: They don't know. And then the ones we're really impressed

357
00:17:44,160 --> 00:17:46,639
Speaker 3: with actually don't know anything that we as as a

358
00:17:46,720 --> 00:17:50,320
Speaker 3: human species don't already know and haven't already thought of

359
00:17:50,359 --> 00:17:52,959
Speaker 3: and written down, because that's all all the information has

360
00:17:52,960 --> 00:17:56,399
Speaker 3: been trained on, is all this human written information, So

361
00:17:56,680 --> 00:17:59,880
Speaker 3: it doesn't have a way of answering beyond is training data.

362
00:18:01,240 --> 00:18:04,520
Speaker 1: Listen to more Coast to Coast AM every weeknight at

363
00:18:04,560 --> 00:18:07,800
Speaker 1: one a m. Eastern and go to Coast tocoastam dot

364
00:18:07,840 --> 00:18:08,600
Speaker 1: com for more